Publication number | US7489797 B2 |
Publication type | Grant |
Application number | US 10/567,735 |
PCT number | PCT/EP2004/051265 |
Publication date | Feb 10, 2009 |
Filing date | Jun 28, 2004 |
Priority date | Aug 8, 2003 |
Fee status | Paid |
Also published as | CA2533907A1, CA2533907C, EP1661084A1, US20070009134, WO2005015493A1 |
Publication number | 10567735, 567735, PCT/2004/51265, PCT/EP/2004/051265, PCT/EP/2004/51265, PCT/EP/4/051265, PCT/EP/4/51265, PCT/EP2004/051265, PCT/EP2004/51265, PCT/EP2004051265, PCT/EP200451265, PCT/EP4/051265, PCT/EP4/51265, PCT/EP4051265, PCT/EP451265, US 7489797 B2, US 7489797B2, US-B2-7489797, US7489797 B2, US7489797B2 |
Inventors | Ebroul Izquierdo |
Original Assignee | Motorola, Inc. |
Export Citation | BiBTeX, EndNote, RefMan |
Patent Citations (13), Non-Patent Citations (9), Referenced by (10), Classifications (26), Legal Events (4) | |
External Links: USPTO, USPTO Assignment, Espacenet | |
This application claims the benefit of prior filed co-pending international application Ser. No. PCT/EP2004/051265 filed on Jun. 28, 2004, and Great Britain application Ser. No. 0318651.7 filed on Aug. 8, 2003. Both of these applications are assigned to Motorola, Inc.
The invention relates to a method and apparatus for fragile watermarking, and in particular a method and apparatus for fragile watermarking and a method for validating such a fragile watermark.
Photographs, paintings, film material and other artistic works have for many years been recorded and transmitted using analogue carriers. However, their reproduction and processing is time consuming, involves a heavy workload and leads to degradation of the original material. This means that content produced and stored using analogue devices has an in-built protection against unintentional changes and malicious manipulation. In general, deliberate changes in analogue media are not only difficult but can easily be perceived by a human inspector.
Recently however, digital media have become pervasive, and threaten to completely substitute their analogue counterparts. Furthermore, affordable media processing tools and fast transmission mechanisms are ubiquitous. As a consequence digital content can nowadays be accurately copied, processed and distributed around the world within seconds. Creators, legitimate distributors and end-users enjoy the flexibility and user friendliness of digital processing tools and networks to copy, process and distribute their content over open digital channels at high speed. However, they also need to guarantee that material used or being published at the end of the distribution chain is genuine. Consequently, automatic tools to establish the authenticity and integrity of digital media are highly important.
Secure communications problems have largely found a solution in cryptography, which guarantees message integrity by using digital signatures with secret keys. However, traditional cryptosystems do not permanently associate cryptographic information with the content. Cryptographic techniques do not embed information directly into the message itself, but rather hide a message during communication.
To provide security by using signatures embedded directly in the content, additional methods need to be considered. Techniques that have been proposed to address this problem belong to a more general class of methods known as digital watermarking, as for example may be found in Signal Processing, Special Issue on Watermarking, vol. 66, no. 3 May 1998.
Several watermarking schemes that address image authentication have been previously developed and fall into two basic categories: fragile and semi-fragile.
Fragile watermarking schemes address the detection of any image changes. Semi-fragile watermarking schemes are designed to discriminate between expected image changes, in most cases due to application constraints, e.g., compression to meet bandwidth requirements, and intentional image tampering.
In the case of fragile watermarking, a number of schemes exist in the prior art:
One prior scheme is proposed in S. Walton, “Information Authentication for a Slippery New Age”, Dr. Dobbs Journal, vol. 20, no. 4, April 1995, pp. 18-26. The scheme uses a check-sum built from the 7 most significant bits of a given pixel, which is then inserted as the least significant bit of the pixel. However, the watermark has only limited security, primarily due to the ease of calculating new check-sums.
Another prior fragile watermarking scheme is proposed in M. M. Yeung and F. Mintzer, “An Invisible Watermarking Technique for Image Verification”, Proc. ICIP, Santa Barbara, Calif., 1997. The Yeung-Mintzer algorithm uses a secret key to generate a unique mapping that randomly assigns a binary value to grey levels of the image. This mapping is used to insert a binary logo or signature in the pixel values. Image integrity is inspected by direct comparison between the inserted logo or signature and the decoded binary image. The main advantage of this algorithm is its high localization accuracy derived from the fact that each pixel is individually watermarked. However, the Yeung-Mintzer algorithm is vulnerable to simple attacks as shown in J. Fridrich, “Security of Fragile Authentication Watermarks with localization”, Proc. SPIE, vol. 4675, No. 75, January 2002.
A third prior scheme for image authentication is proposed in P. W. Wong, “A Public Key Watermark for Image Verification and Authentication”, Proc. ICIP, Chicago, Ill., October 1998. This scheme embeds a digital signature extracted from the most significant bits of a block of the image into the least significant bit of the pixels in the same block. However this scheme was shown to be vulnerable to a counterfeiting attack in M. Holliman and N. Memon, “Counterfeiting Attacks on Oblivious Block-Wise Independent Invisible Watermarking Schemes”, Proc. IEEE Trans. on Image Processing, vol 9, no 3, March 2000, pp. 432-441. This attack belongs to the class of vector quantization counterfeiting and has been shown to defeat any fragile watermarking scheme that achieves localization accuracy by watermarking small independent image blocks.
One common feature of these and other prior schemes from the literature is that authentication signatures are embedded in the image content, either in the pixel or a transform domain, and the security of the schemes resides in a hash or encryption mechanism.
This ultimately leaves such schemes vulnerable to the attacks noted above.
Thus there is a need for an alternative method of fragile watermarking.
The purpose of the present invention is to address the above problem.
The present invention provides a method of fragile watermarking, characterised by the step of generating at least a first ill-conditioned operator, said ill-conditioned operator being related to values extracted from an image or portion thereof A.
In a first aspect, the present invention provides a method of fragile watermarking, as claimed in claim 1.
In a second aspect, the present invention provides a method of verifying a fragile watermark.
In a third aspect, the present invention provides apparatus for fragile watermarking.
In a fourth aspect, the present invention provides apparatus for verifying a fragile watermark.
Further features of the present invention are as defined in the dependent claims.
Embodiments of the present invention will now be described by way of example with reference to the accompanying drawings, in which:
Referring to
An embodiment of the present invention provides a method providing an essentially different approach from those reported in the prior art. This method is based on the inherent instability property of inverse ill-conditioned problems, and the fact that small changes to their input data cause large changes in any approximate solution. Singular valued decomposition and other fundamental linear algebra tools are used to construct an ill-conditioned matrix interrelating the original image and the watermark pattern. This is achieved by exploiting the relationship between singular values, the least square solution of linear algebraic equations and the high instability of linear ill-conditioned operators.
In brief, an embodiment of the present invention performs a fragile watermarking method on blocks (portions) of pixels extracted from the image to be watermarked, the values in this block being treated as a matrix for the purposes of analysis.
Similarly, blocks of pixels of corresponding size are taken from a watermark pattern, or are generated to resemble such blocks.
The smallest singular value of the matrix of the watermark is replaced to artificially create an ill-conditioned minimization problem. The solution to this problem involves a least squares approximation of the previously defined ill-conditioned operator in order to find an unknown parameter.
This solution process links the watermark with the image using the underlying ill-conditioned operator. An image block is considered watermarked by setting its smallest singular value equal to a parameter estimated from the minimization task.
Thus, the watermark is spread over the whole image in a subtle but quite complex manner. One advantage of this method is that the distortion induced by the watermarking procedure can be strictly controlled since it depends only on changes in the smallest singular values of each block.
The verification procedure solves the same optimisation problem and compares the norm of the solution with a large secret number N used in the generation of the watermark.
Any small change to the image will therefore result in the norm of the solution differing significantly from the large secret number N, due to the property of the ill conditioned operator to produce highly differing solutions in response to small changes in the input values.
By contrast, known methods of watermarking have considered the presence of ill-conditioned operators to be an unintended nuisance that must be overcome, such as U.S. Pat. No. 6,282,300 (Bloom), e.g. during watermarking validation.
The method will now be described in detail;
Firstly, to illustrate how an ill-conditioned operator may be used to provide a watermark, consider that in many known applications of linear algebra, it is necessary to find a good approximation {circumflex over (x)} of an unknown vector xε
^{n }satisfying the linear equationHowever, if B is ill-conditioned or singular then {circumflex over (x)}=B^{+}b, if it exist at all, is a poor approximation of x.
An error estimate given by ∥x−{circumflex over (x)}∥≦∥B^{+}∥∥B{circumflex over (x)}−b∥ shows that the approximation error can grow proportional to the norm of the inverse of B. Since the norm of the inverse of B is proportional to the condition number of B, it is evident that the more the ill-conditioning of B, the larger the different between x and {circumflex over (x)}.
Furthermore, the estimation of the inverse of an ill-conditioned matrix is not straightforward, and clearly is essentially the same problem as seen in Eq. 1. Moreover, when B is ill-conditioned, solving Eq. 1 becomes equivalent to solving the optimisation problem
for a predefined norm ∥.∥. It is well-known in the art that the L_{2}-norm solution of this least squares problem is given by
It becomes evident from Eq. 2 that errors in either any of the left singular vectors of B or in the right-hand side b are drastically magnified by the smallest of the singular values S_{i}(B) of B.
In an embodiment of the present invention, an ill-conditioned operator B is generated that is related to values extracted from an image or a portion thereof A. In this manner alterations to A will be reflected in magnified errors to a solution of {circumflex over (x)} of similar form to Eq. 2.
In an embodiment of the present invention, a watermarking pattern Ω is interrelated with an image I, thus watermarking it, as follows:
Given an image I of dimensions m×n, a watermark pattern Ω of typically the same dimensions is built.
In a first embodiment, Ω is an array of pseudo-randomly generated binary or real numbers.
In an alternative embodiment of the present invention, the procedure to generate Ω uses a single or repeated instance of a logo, typically a binary pattern, combined with pseudo-randomly generated numbers; initially, a mosaic-like binary image P of dimension m×n is built by tiling the logo to occupy an area similar to the original image I. The watermark pattern is then typically defined as Ω=P⊕ω, where ω is a m×n array of pseudo-randomly generated binary numbers and ⊕ denotes the bitwise XOR operator.
In either embodiment, no assumption needs to be imposed on the statistical properties of the random number generator; the binary or real numbers used to generate the watermark can follow any probability distribution and are not restricted to Gaussian or uniform. This is because the present invention does not rely on statistical analysis for authentication or tamper detection. However it will be clear to a person skilled in the art that in consequence statistical constraints can be imposed if desired.
In either embodiment, Ω depends on a secret key K whose value seeds the pseudo-random number generator. K is subsequently also used in validating the watermark.
In an embodiment of the present invention, the watermarking process is performed in a block-wise fashion. For the sake of simplicity and without loss of generality, for the following description assume that an image I is partitioned into L small blocks or portions A^{(k)}, k=1, . . . , L of dimensions p×q. Likewise, Ω is partitioned into L blocks or portions W^{(k)}, k=1, . . . , L of dimensions p×q. For the sake of notational simplicity the upper index representing the block number will be omitted hereon in unless expressly referred to. Without loss of generality for the following description assume that the blocks are square, i.e., p=q=n.
Thus blocks A and W can be considered for the purpose of explanation to be generic matrices comprising values obtained from the original image and watermark, respectively.
As is known in the art, a fundamental result of Linear Algebra states that matrix A can be represented as
A=U _{A} S _{A} V _{A} ^{T}, Eq.3
i.e. a singular value decomposition of A, where U_{A}=(u_{1}, . . . , u_{n})ε
In an embodiment of the present invention, it is proposed to replace S_{r}(A) with a real positive number ŝ_{r}(A) as part of the process to produce a watermarked version Â of A, where the distortion introduced by the watermarking process is determined with reference to the calculation of
∥A−Â∥ _{2} =|s _{r}(A)−ŝ _{r}(A), Eq. 4
where ∥.∥_{2 }denotes the L_{2}-norm, and thus the distortion is dependent upon the value of ŝ_{r}(A).
Given an image or portion thereof A, the corresponding watermarked portion or block is defined as the matrix Â, generated according to the following considerations. Observe that A and Â have the same dimensions.
Initially, singular value decomposition of A and W is performed to obtain A=U_{A}S_{A}V_{A} ^{T }and W=U_{W}S_{W}V_{W} ^{T}, respectively. Let S_{A}=diag(s_{1}(A), . . . , s_{r}(A)) and S_{W}=diag(s_{1}(W), . . . , s_{t}(W)) be the nonzero singular values of A and W respectively. The two diagonal matrices Ŝ_{A} 32 diag(s_{1}(A), . . . , ŝ_{r}(A)) and Ŝ_{W}=diag(s_{1}(W), . . . , ŝ_{t}(W)) are then built by replacing the last nonzero singular values s_{r}(A) and s_{t}(W) by two specific real positive numbers ŝ_{r}(A) and ŝ_{t}(W), respectively. Here it is assumed that the smallest nonzero singular value of A is s_{r}(A), i.e., rank(A)=r and the smallest nonzero singular value of W is s_{t}(W), i.e., rank(W)=t. Using Ŝ_{A }the watermarked block Â is defined as
Â=U _{A} Ŝ _{A} V _{A} ^{T}. Eq. 5
Likewise, Ŝ_{w }is used to build an ill-conditioned matrix Ŵ according to
Ŵ=U _{W} Ŝ _{W} V _{W} ^{T}. Eq. 6
Now, one should choose the two values ŝ_{r}(A) and ŝ_{t}(W).
In selecting values of ŝ_{r}(A) and ŝ_{t}(W), it is desirable to do so in such a fashion as to facilitate the fragility of the watermarking process, the uniqueness of the watermark thus made and optionally the control of perceptibility of the watermark in the final watermarked image Î.
A. Fragility
It is desired that any change to single or multiple elements of Â can be detected by a validation procedure.
In an embodiment of the present invention, replacing s_{t}(W) with ε in the calculation of Eq. 6 achieves this if ε is a sufficiently small positive real number, increasing the condition number of the singular value matrix S_{W }and so making Ŵ extremely ill-conditioned. Â and Ŵ are then interrelated using matrix multiplication to produce the ill conditioned matrix B=ÂŴ.
Although by addressing the requirement of fragility Ŵ is now defined using ε, Â still depends on an unknown parameter ŝ_{r}(A). For that reason B should be regarded as a parametric family of matrices:
B(ŝ _{r})=Â(ŝ _{r})Ŵ. Eq. 7
This parametric family of matrices B(ŝ_{r}) 110 determines the linear ill-conditioned operator used in the fragile watermarking method, and is resolved by addressing the second requirement:
B. Uniqueness
It is desirable to select from amongst the parametric family of matrices B(ŝ_{r}) a single operator for use in a specific fragile watermark.
For a pre-defined large real number N, there exists a unique value of ŝ_{r}(A), so that the L_{2}-norm solution of the least squares problem
is N^{2}. Here, b is an arbitrary vector defining the right-hand side of the linear system to be minimized in Eq. 8.
Thus in an embodiment of the present invention, by selecting a value of N as a key, a corresponding unique value
By using this unique value
C. Perceptibility
Whilst the processes described above to address the conditions of fragility and uniqueness are sufficient to provide a watermarked block Â, in an enhanced embodiment of the present invention the selected value of ŝ_{r}(A) is additionally constrained to lie in the interval max(eps, s_{r}(A)−δ)≦ŝ_{r}(A)≦s_{r}(A)+δ, where eps is the machine precision and δ is a scalar used to control the distortion to the image block A induced by the watermark in Â.
The expression max(eps, s_{r}(A)−δ) ensures that ŝ_{r}(A) remains nonzero and positive. This condition together with Eq. 2 allows the distortion to be kept below a user-defined value δ.
In an embodiment of the present invention, the method of fragile watermarking of an image I comprises the following steps:
(based on Eq. 2) where u_{B} _{ i }is the i-th column of the matrix formed with the right singular vectors of B, s_{i}(B) are the singular values of B, b is the right-hand side vector given in Eq. 8 and key N is a large real number.
In an otherwise similar enhanced embodiment of the present invention, step iii. above comprises estimating the unique parameter
In both the directly preceding embodiments, step iv. shows how the value ŝ_{r}(A) in Eq. 5 is chosen, namely by setting ŝ_{r}(A)=
In an enhanced embodiment of the present invention, the value of b selected for equations 8, 9 or 10 is made dependant upon a parameter derived from a portion of image I other than current portion A:
For a sequential watermarking process comprising the watermarking of portion A^{(k) }after the watermarking of portion A^{(k−1)}, for k=1, . . . , L of L portions of image I, then the step of calculating b^{(k) }for portion A^{(k) }comprises calculating substantially the following equation part:
where Z^{(k) }is a pseudo-random binary vector.
This enhancement increases the difficulty of successfully undertaking a vector quantisation attack upon the image I, requiring that larger image areas containing several authenticated blocks are replaced. Even then, the blocks at the border of the swapped area will be declared faked:
To validate authenticity and to detect tampered areas, a receiver of a received image I′ needs to test if the received image or a portion thereof A′ has been tampered with or not. It is assumed that the receiver is a trusted party who knows the secret set of keys κ={K, N}.
In an embodiment of the present invention, in addition to ε a tolerance value τ is used in the verification process. This parameter provides tolerance to approximation errors inherent to any numerical process. ε and τ are fixed numbers and so can be known to the public. Referring to
Using K, the receiver first generates the watermark pattern or portion thereof W. Next, ε is used to build the matrix Ŵ by setting Ŝ_{w}=diag(s_{1}(W), . . . , ε) as in Eq. 6. Afterwards, the ill-conditioned matrix B′=A′Ŵ is built 210 and the solution of the minimization problem
is calculated 220. Once Eq. 12 has been solved N′ is defined as the square root of the norm of the vector x minimizing Eq. 12.
The verification step consists of a comparison between N′ and the secret value N 230. A Boolean response is obtained by thresholding the absolute difference |N′−N|=γ. If γ≦τ, A′ is authentic 232, otherwise A′ is declared a fake 234, as it is judged that modifications to A′ have altered the ill-conditioned matrix B^{+}=A′Ŵ such that the error in solution N′ to expected solution N exceeds tolerance threshold τ.
It will be clear to a person skilled in the art that whilst ŝ_{r}(A) and ŝ_{t}(W) are the preferred singular values to be replaced, an embodiment of the present invention may replace a singular value other than ŝ_{r}(A) or ŝ_{t}(W), although for ŝ_{r}(A) this is likely to increase distortion in the watermarked block Â.
It will also be clear to a person skilled in the art that tractable linear and non-linear problems other than the minimisation problem of Eq. 8 and 12 that involve an ill-conditioned operator may be amenable to the methods described herein.
For the purposes of clarity, the following provides detailed proofs of the ability to find an ill-conditioned operator B for a given A, and the ability to find a value
To prove the ill-conditioning of B, let A and W be two square matrices of the same dimension and s_{k}(A), s_{k}(W) their k-th singular values, respectively. Then, s_{i+j−1}(AW)≦s_{i}(A)s_{j}(W), for all integers i, j. (For the proof of this result, see A. Pietsch, Eigenvalues and s-Numbers, Cambridge University Press, 1997, Proposition 2.3.12.)
Next, let the smallest singular values of B=AW and W be s_{r}(B) and s_{t}(W), respectively. Then
s _{r}(B)≦s _{r−t+1}(A)·s _{t}(W)=ε·s_{r−t+1}(A) for t≦r. Eq. 13
This follows directly from the previous result by setting i=r−t+1 and j=t.
Since ε is chosen to be very small, the inequality Eq. 13 guarantees that the smallest singular value of B is also tiny and therefore extremely ill-conditioned.
Usually, the matrices A and W have full rank, i.e., t=r. However, it is possible to build counterexamples with t>r. Even in such usual situations Eq. 13 can be applied by setting s_{k}(W)=0 for all k>r. Observe that because Ŵ is artificially constructed, there is nothing to prevents the required values being set to zero. As a consequent the condition t≦r in Eq. 13 can be assumed in any case.
In order to provide the existence of
h(z) can be written as h(z)=s_{r}(A(z)Ŵ)≡(h_{1}∘h_{2})(z), with h_{1}(z)=s_{r}(B(z)) and h_{2}(z)=A(z)Ŵ. The two functions h_{1 }and h_{2 }are continuous in the interval [H_{0}, H_{1}]. Hence, h(z) is also continuous in [H_{0}, H_{1}]. The continuity of h(z) can now be used to prove that g(z) is continuous in (H_{0}, H_{1}]. Using Eq. 2 it is straightforward to derive the following expression:
Thus, g(z) is the sum of quotients of continuous functions. Therefore, g(z) is also continuous in [H_{0}, H_{1}].
Now, consider h max=max(g(z)) and h min=min(g(z)). If Nε[g(h max), g(h min)] then it exists
The above considerations illustrate the effectiveness and feasibility of the proposed invention. The underlying operator of Eq. 8 can be made extremely ill-conditioned while the norm of its solution is kept equal to N. Furthermore, by selecting ŝ(A)ε[H_{0}, H_{1}] the distortion on the original image remains below the input parameter δ.
However, this last property constrains the variation of ŝ_{r}(A) to a very small interval. Since ŝ_{r}(A) depends on N, an important question arises of how the small interval [H_{0}, H_{1}] constrains the set of feasible values N.
Since N is a secret key it is desirable that it is extremely difficult to estimate. Obviously the smaller the set of feasible values for N, the easier it is to estimate N and so mount a successful attack. This concern is addressed below.
Fortunately, the range of values that can be used for N is large, making difficult for an attacker to estimate it. Since the distortion introduced by the watermark can be strictly controlled by the distortion coefficient δ, this coefficient defines the feasibility interval [H_{0}, H_{1}]. Clearly, this interval is very small. Its maximum length does not exceed 2δ and according to the considerations above it defines the range of permissible values for Nε[g(h max), g(h min)]. Since N should be a large number to improve the security of the proposed algorithm, it is also important to show that the interval of permissible values of N is also very large. Variations of zε[H_{0}, H_{1}] are reflected in the variations of the smallest singular value of B. According to Eq. 13 the smallest singular value of B is very closet to ε. This fact can be used to find an estimate for the interval [g(h max), g(h min)]. For this we consider the hyperbola p(z)=C+D/y^{2 }with
and D=(u_{Hr} ^{T}b)^{2}. Since the variation of zε[H_{0}, H_{1}] determines the variation of p(z), this gives the range for possible values N. Observe that changes in z also affect C and D, but actually the smallest singular value of B is the leading term determining the behaviour of p(z). Clearly, p(z)→∞ if z→0. Furthermore, p maps tiny intervals very close to zero into very large intervals. For instance, if ε=10^{−16 }and δ=10^{−2}, then z will approximately vary between the machine precision eps, e.g., 10^{−32}, and 10^{−2}. In this case [g(h max), g(h min)]≈[10^{2}, 10 ^{32}]. As a consequence, for this particular example N could be selected from the interval Nε[10^{2}, 10^{32}]. These arguments show that the range of permissible values of N is huge and it would be extremely hard for an attacker to estimate N.
In this specification, the expression ‘condition number’ (or ‘matrix condition number’) is referred to. This expression is well known in the field of matrix computations. The condition number
κ(A)
of a square matrix A is defined as
κ(A)=∥A∥∥A ^{−1}∥
where
The (matrix) condition number is basically a measure of stability or sensitivity of a matrix (or the linear system it represents) to numerical operations. In other words, we may not be able to trust the results of computations on an ill-conditioned matrix. Matrices with condition numbers near 1 are said to be well-conditioned. Matrices with condition numbers much greater than one, e.g. 10^{n }for an n-sided matrix (such as around 10^{5 }for a 5×5 Hilbert matrix), are said to be ill-conditioned. Thus, a condition number less than 5, preferably near to 1, can be considered to give a well conditioned matrix and condition numbers greater than 50, preferably about 100 or more, can be considered to give an ill-conditioned matrix.
Cited Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|
US5960081 * | Jun 5, 1997 | Sep 28, 1999 | Cray Research, Inc. | Embedding a digital signature in a video sequence |
US6064764 * | Mar 30, 1998 | May 16, 2000 | Seiko Epson Corporation | Fragile watermarks for detecting tampering in images |
US6633653 | Feb 4, 2000 | Oct 14, 2003 | Motorola, Inc. | Watermarked digital images |
US7061510 * | Mar 5, 2001 | Jun 13, 2006 | Digimarc Corporation | Geo-referencing of aerial imagery using embedded image identifiers and cross-referenced data sets |
US7401048 * | May 30, 2002 | Jul 15, 2008 | Jpmorgan Chase Bank, N.A. | System and method for trade settlement tracking and relative ranking |
US20020178368 | May 24, 2001 | Nov 28, 2002 | Peng Yin | Semi-fragile watermarking system for MPEG video authentication |
US20030070075 | Jul 15, 2002 | Apr 10, 2003 | Frederic Deguillaume | Secure hybrid robust watermarking resistant against tampering and copy-attack |
US20050144454 | Jun 17, 2002 | Jun 30, 2005 | Hare Jonathan S. | Video/image communication with watermarking |
US20070172094 * | Apr 4, 2003 | Jul 26, 2007 | Datamark Technologies Pte Ltd | Watermarking method and apparatus |
US20070223778 | Jun 17, 2002 | Sep 27, 2007 | Paola Hobson | Method And Apparatus For Video/Image Communication With Watermarking |
EP0947953A2 | Mar 16, 1999 | Oct 6, 1999 | Seiko Epson Corporation | Watermarks for detecting tampering in images |
GB2370437A | Title not available | |||
GB2377575A | Title not available |
Reference | ||
---|---|---|
1 | Chandra, "Digital Image Watermarking Using singular Value Decomposition", 2002 45th Midwest Symposium on Circuits and Systems. Conference Proceedings (Cat. No. 02CH37378) IEEE Piscataway, NJ USA, vol. 3, 2002, pp. III-24-III-267, XP002300902; ISBN: 0-7803-7523-8, Section 3. | |
2 | * | Digital Image watermarking using Singular value decomposition, Chandra et al. |
3 | Gorodetski, et al., "SVD-Based Approach to Transparent Embedding Data into digital Images", Information Assurance in Computer Networks, Methods, Models and Architecture for Network Security. International workshop MMM-ACNS 2001. Proceedings (Lecture Notes in Computer Science vol. 2052) Springer-Verlag Berlin, Germany, 2001, pp. 263-274, XP002300903, ISBN: 3-450-42103-3, Section 3. | |
4 | Lei, et al., "Using III-posed Matrix in Correlation-based Digital Watermarking", Proceedings of the SPIE-the International Society for Optical Engineering SPIE-Int. Soc. Opt. Eng USA, vol. 4861, 2002, pp. 183-191, XP002300904, ISSN: 0277-786X: p. 184-185. | |
5 | Rui Sun, et al., "A SVD-and Quantization Based Semi-Fragile Watermarking Technique for Image Authentication", Proceedings of the ICSP, vol. 2, Aug. 26, 2002 pp. 1592-1595, XP010627843, Abstract, Section 3. | |
6 | Ruizhen, et al., "An SVD-based Watermarking Scheme for Protecting Rightful Ownership", IEEE Transactions on Multimedia IEEE USA, vol. 4, No. 1, Mar. 2002 (2002-2003), pp. 121-128, XP008036943, ISSN: 1520-9210, Section III B. | |
7 | * | Sun et al., A SVD and quantization based semi-fragile watermarking technique for image authentication. Aug. 2003. |
8 | * | SVD Based Approach to transport embedding data into digital images, Gorodetski et al. |
9 | * | Wong, A public key watermark image verification and authentication, 1998. |
Citing Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|
US7787654 * | Sep 21, 2007 | Aug 31, 2010 | Shibaura Institute Of Technology | Electronic watermark embedding apparatus and method and electronic watermark detecting apparatus and method |
US7881492 | Dec 14, 2007 | Feb 1, 2011 | Shibaura Institute Of Technology | Electronic watermark embedding apparatus and method and electronic watermark examining apparatus and method |
US7930546 * | Jan 2, 2007 | Apr 19, 2011 | Digimarc Corporation | Methods, systems, and sub-combinations useful in media identification |
US7965838 * | Nov 30, 2007 | Jun 21, 2011 | Faraday Technology Corp. | Watermark generating circuit and related method thereof |
US8712738 | Apr 30, 2010 | Apr 29, 2014 | International Business Machines Corporation | Determining ill conditioning in square linear system of equations |
US20050147248 * | Feb 26, 2003 | Jul 7, 2005 | Koninklijke Philips Electronics N.V. | Window shaping functions for watermarking of multimedia signals |
US20070174059 * | Jan 2, 2007 | Jul 26, 2007 | Rhoads Geoffrey B | Methods, Systems, and Sub-Combinations Useful in Media Identification |
US20090141927 * | Nov 30, 2007 | Jun 4, 2009 | Yu-Min Wang | Watermark generating circuit and related method thereof |
US20100183190 * | Sep 21, 2007 | Jul 22, 2010 | Kazuo Ohzeki | Electronic watermark embedding apparatus and method and electronic watermark detecting apparatus and method |
US20100202651 * | Dec 14, 2007 | Aug 12, 2010 | Shibaura Institute Of Technology | Electronic watermark embedding apparatus and method and electronic watermark examining apparatus and method |
U.S. Classification | 382/100, 283/902, 380/201, 370/529, 380/51, 382/240, 382/232, 380/54, 380/210, 370/522, 358/3.28, 283/72, 380/252, 283/901, 283/113, 713/179, 713/176 |
International Classification | G06T1/00, H04N1/32 |
Cooperative Classification | Y10S283/901, G06T1/0042, G06T2201/0061, G06T2201/0051, G06T2201/0052, Y10S283/902 |
European Classification | G06T1/00W4 |
Date | Code | Event | Description |
---|---|---|---|
Jul 26, 2006 | AS | Assignment | Owner name: MOTOROLA, INC., ILLINOIS Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:IZQUIERDO, EBROUL;REEL/FRAME:018006/0437 Effective date: 20040510 |
Apr 6, 2011 | AS | Assignment | Owner name: MOTOROLA SOLUTIONS, INC., ILLINOIS Free format text: CHANGE OF NAME;ASSIGNOR:MOTOROLA, INC;REEL/FRAME:026081/0001 Effective date: 20110104 |
Jul 25, 2012 | FPAY | Fee payment | Year of fee payment: 4 |
Jul 25, 2016 | FPAY | Fee payment | Year of fee payment: 8 |